In patients' words: natural language processing of reports from patients experiencing orofacial pain and dysfunction.

backgroundProvision of value-based, patient-centered care requires careful appraisal of patients’ symptoms. Understanding subjective experiences, needs, and feelings is crucial for shared decision-making, especially when clinical findings differ from individual perceptions. Natural language processing (NLP) offers new ways to understand patients’ perspectives. This exploratory pilot study aimed to exemplify the use of three popular NLP methods to analyze open-ended textual self-reports from individuals experiencing orofacial pain and dysfunction for a more comprehensive understanding of their subjective symptom burdens.

methodsThis study used topic modeling, conceptual maps, and lexicon-based linguistic style analysis to analyze texts from 2,237 patients experiencing orofacial pain and/or dysfunction, who provided brief written descriptions of their chief complaints, functional limitations, and expectations.

resultsFrom the aggregated text corpus of 111,923 words, unsupervised topic modeling identified 10 meaningful topics by clustering words related to prevalent complaints, constraints, and co-morbidities like tinnitus and insomnia, highlighting patients’ hopes for understanding causes and receiving a clear diagnosis. Conceptual maps of the 200 most frequent words or expressions revealed occupational limitations as significant constraints and highlighted the patients’ need for understanding causes and diagnoses. Linguistic style analyses were used to enrich the map, revealing negative emotional associations with chief complaints and the patients’ struggle to reduce uncertainty and understand their illness. DISCUSSION: The results revealed distinct language patterns in open-ended orofacial pain reports. Chief complaints were associated with terms linked to anatomical locations and temporal patterns, functional limitations with impaired masticatory function, work-related activities and sleep disturbances, and expectations with an improved understanding of symptoms. Adding linguistic categories allowed for the validation of unsupervised methods and offered a nuanced approach to evaluate symptom burdens. NLP methods complement traditional information collection by capturing patients’ views, which are crucial for healthcare practice and shared decision-making within a biopsychosocial framework. When integrated into clinical workflows, NLP technologies might be a promising way of enhancing comprehensive symptom appraisal, benefiting both patients and clinicians alike.